Related Experiment Video
Updated: Jun 28, 2026
16:14
Trajectory Data Analyses for Pedestrian Space-time Activity Study
Published on: February 25, 2013
13.4K
Comparing XAI techniques for interpreting short-term burglary predictions at micro-places.
Robin Khalfa1, Naomi Theinert1, Wim Hardyns1,2
1Department of Criminology, Criminal Law and Social Law, Ghent University, Universiteitstraat 4, Ghent, 9000 Belgium.
Computational Urban Science
|May 12, 2025
Summary
This study compares explainable AI (XAI) methods for interpreting burglary predictions. It finds built environment features are key global predictors, but local explanations vary, urging careful method selection for crime prevention.
Area of Science:
- Criminology
- Computer Science
- Artificial Intelligence
Background:
- Machine learning models are increasingly used for crime prediction.
- Existing research often uses SHapley Additive exPlanations (SHAP) for interpreting these models.
- There is a need to compare various eXplainable Artificial Intelligence (XAI) techniques for crime prediction.
Purpose of the Study:
- To empirically compare multiple XAI techniques for interpreting short-term burglary predictions.
- To evaluate SHAP alongside other XAI methods for both global and local interpretability.
- To assess the impact of different XAI methods on crime prevention strategies.
Main Methods:
- Trained an XGBoost model using 76 features (2014-2018 data) to predict weekly residential burglary hot spots in Ghent, Belgium.
- Systematically evaluated SHAP, LIME, and other XAI techniques for model interpretation.
- Analyzed global and local feature importance, focusing on built environment, land use, socio-demographics, and seasonality.
Main Results:
- Built environment and land use were consistent global predictors of burglary risk.
- Local explanations for feature importance varied significantly across XAI techniques, particularly between SHAP and LIME.
- Complex interactions between environmental and social disorganization features influenced short-term burglary risk.
Conclusions:
- The choice of XAI technique is critical when translating machine learning predictions into actionable crime prevention strategies.
- Integrated approaches are needed to understand the complex interplay of crime predictors.
- Further attention to the methodological implications of XAI in crime science is warranted.
